Foundation models, such as GPT-4, are ushering in a new era of AI by outperforming humans on numerous tasks across modalities including images and language. With the introduction of the Foundry add-on for Labelbox Model, we’re bringing the power of foundation models into the Labelbox platform.
Meet your new AI co-pilot: Foundry
Labelbox Foundry enables AI teams to use world-class foundation models to enrich datasets and automate tasks. In just a few clicks, AI builders can explore, test and integrate powerful models to build vital workflows for pre-labeling, or other specific data tasks. Kickstart your AI efforts with this AI copilot to build intelligent applications faster than ever.
Early tests with Fortune 500 companies show up to 88% reductions in human labeling time, with complex tasks going from days to hours. Foundry integrates these state-of-the-art models into every step of the workflow, unlocking the next evolution of AI development. Kickstart your AI and data labeling efforts with this co-pilot to build intelligent applications faster than ever.
We’re excited to announce that Foundry is available as an add-on for Labelbox Model!
With Foundry, you can:
Foundry works across Labelbox's products — Catalog, Annotate, and Model — to supercharge your labeling and model development.
Foundry is only available to our Starter and Enterprise plans, to access Foundry you will need to:
1) Upgrade your account to our Starter plan:
2) After upgrading to our Starter plan, please follow the instructions under the Starter plan below to activate Foundry as an add-on for your organization.
As part of the self-serve Starter or Standard plan, you automatically have the ability to opt-in to using Foundry.
To generate model predictions, just submit a model run and you’ll be guided to activate Foundry along the way:
1) Generate predictions: Start by selecting data in Catalog and click “Predict with Foundry.” You’ll be able to select a foundation model and submit a model run to generate predictions.
2) Activate Foundry: The first time you submit a model run, you’ll see the option to activate Foundry for your organization’s Admins. You’ll need to agree to Labelbox’s Model Foundry add-on service terms and confirm you understand the associated compute fees.
3) Payment method: With Foundry, you will be charged for the Labelbox units (LBUs) your account consumes, plus any pass-through compute costs for the models you use.
For self-serve Starter and Standard tier to enable billing for these model compute charges, you'll be asked to confirm the credit card on file for your Labelbox account. Alternatively, you may add another card if you prefer to keep the Foundry charges separate. By confirming your payment method, you agree to let Labelbox to bill your card as Foundry model compute fees accrue based on your usage.
As an organization on our Enterprise plan, you automatically have the ability to opt-in to using Foundry.
To generate model predictions, just submit a model run and you’ll be guided to activate Foundry along the way:
1) Generate predictions: Start by selecting data in Catalog and click “Predict with Foundry.” You’ll be able to select a foundation model and submit a model run to generate predictions.
2) Activate Foundry: The first time you submit a model run, you’ll see the option to activate Foundry for your organization’s Admins. You’ll need to agree to Labelbox’s Model Foundry add-on service terms and confirm you understand the associated compute fees.
3) Payment method: With Foundry, you will be charged for the Labelbox units (LBUs) your account consumes, plus any pass-through compute costs for the models you use.
If you are on an annual contract plan, you may add a credit card on file to pay for the associated compute costs of a model or choose to receive a monthly invoice at the end of each month based on your organization’s Foundry usage.
In the below interactive demos, you'll learn how to leverage foundation models to generate model predictions, send predictions as pre-labels to a labeling project in Annotate, and verify the predictions with human-in-the-loop review:
Access a vast range of ready-to-use foundation models that embed advanced AI into your data tasks with ease. Quickly generate predictions to pre-label datasets or to enrich your existing data to extract better insights that boost productivity and increase time-savings.
Rather than labeling from scratch, combine the power of foundation models with human-in-the-loop review to accelerate your labeling operations.
Focus human intelligence on critical quality assurance instead of on initial labeling efforts. Seamlessly validate model-generated pre-labels in Annotate – approving accurate predictions with a click and easily editing or sending incorrect labels to be corrected.
Check out the below tutorials to learn how Foundry can accelerate your pre-labeling and data enrichment workflows in Labelbox.
Foundry currently supports a variety of tasks for computer vision and natural language processing. This includes:
Watch the below demo to see how to use Foundry for an object detection use case with Amazon Rekognition and Grounding Dino:
This demo uses Amazon Rekognition and Grounding Dino, but you can leverage any other computer vision model available in Foundry.
Watch the below demo to see how to use Foundry for a text classification use case with GPT-4:
This demo uses GPT-4, but you can leverage any other large language model available in Foundry.
Foundry pricing will be calculated and billed monthly based on the following:
Model LBU is consumed for every data row that goes through Foundry. Annotate LBU is consumed for every data row that has a prediction (from Foundry) submitted as a label.
The unit compute costs are listed on the model card for each model found in the Labelbox app.
Learn more about the pricing of the Foundry add-on for Labelbox Model on our pricing page.
If you have questions or feedback related to the Foundry workflow, please submit a ticket here.
To learn more about Foundry and familiarize yourself with the workflow, we recommend checking out the below resources: